1150
1167
# The old API returned a list, should this actually be a set?
1151
1168
return parent_map.keys()
1170
def _check_inventories(self, checker):
1171
"""Check the inventories found from the revision scan.
1173
This is responsible for verifying the sha1 of inventories and
1174
creating a pending_keys set that covers data referenced by inventories.
1176
bar = ui.ui_factory.nested_progress_bar()
1178
self._do_check_inventories(checker, bar)
1182
def _do_check_inventories(self, checker, bar):
1183
"""Helper for _check_inventories."""
1185
keys = {'chk_bytes':set(), 'inventories':set(), 'texts':set()}
1186
kinds = ['chk_bytes', 'texts']
1187
count = len(checker.pending_keys)
1188
bar.update("inventories", 0, 2)
1189
current_keys = checker.pending_keys
1190
checker.pending_keys = {}
1191
# Accumulate current checks.
1192
for key in current_keys:
1193
if key[0] != 'inventories' and key[0] not in kinds:
1194
checker._report_items.append('unknown key type %r' % (key,))
1195
keys[key[0]].add(key[1:])
1196
if keys['inventories']:
1197
# NB: output order *should* be roughly sorted - topo or
1198
# inverse topo depending on repository - either way decent
1199
# to just delta against. However, pre-CHK formats didn't
1200
# try to optimise inventory layout on disk. As such the
1201
# pre-CHK code path does not use inventory deltas.
1203
for record in self.inventories.check(keys=keys['inventories']):
1204
if record.storage_kind == 'absent':
1205
checker._report_items.append(
1206
'Missing inventory {%s}' % (record.key,))
1208
last_object = self._check_record('inventories', record,
1209
checker, last_object,
1210
current_keys[('inventories',) + record.key])
1211
del keys['inventories']
1214
bar.update("texts", 1)
1215
while (checker.pending_keys or keys['chk_bytes']
1217
# Something to check.
1218
current_keys = checker.pending_keys
1219
checker.pending_keys = {}
1220
# Accumulate current checks.
1221
for key in current_keys:
1222
if key[0] not in kinds:
1223
checker._report_items.append('unknown key type %r' % (key,))
1224
keys[key[0]].add(key[1:])
1225
# Check the outermost kind only - inventories || chk_bytes || texts
1229
for record in getattr(self, kind).check(keys=keys[kind]):
1230
if record.storage_kind == 'absent':
1231
checker._report_items.append(
1232
'Missing %s {%s}' % (kind, record.key,))
1234
last_object = self._check_record(kind, record,
1235
checker, last_object, current_keys[(kind,) + record.key])
1239
def _check_record(self, kind, record, checker, last_object, item_data):
1240
"""Check a single text from this repository."""
1241
if kind == 'inventories':
1242
rev_id = record.key[0]
1243
inv = self.deserialise_inventory(rev_id,
1244
record.get_bytes_as('fulltext'))
1245
if last_object is not None:
1246
delta = inv._make_delta(last_object)
1247
for old_path, path, file_id, ie in delta:
1250
ie.check(checker, rev_id, inv)
1252
for path, ie in inv.iter_entries():
1253
ie.check(checker, rev_id, inv)
1254
if self._format.fast_deltas:
1256
elif kind == 'chk_bytes':
1257
# No code written to check chk_bytes for this repo format.
1258
checker._report_items.append(
1259
'unsupported key type chk_bytes for %s' % (record.key,))
1260
elif kind == 'texts':
1261
self._check_text(record, checker, item_data)
1263
checker._report_items.append(
1264
'unknown key type %s for %s' % (kind, record.key))
1266
def _check_text(self, record, checker, item_data):
1267
"""Check a single text."""
1268
# Check it is extractable.
1269
# TODO: check length.
1270
if record.storage_kind == 'chunked':
1271
chunks = record.get_bytes_as(record.storage_kind)
1272
sha1 = osutils.sha_strings(chunks)
1273
length = sum(map(len, chunks))
1275
content = record.get_bytes_as('fulltext')
1276
sha1 = osutils.sha_string(content)
1277
length = len(content)
1278
if item_data and sha1 != item_data[1]:
1279
checker._report_items.append(
1280
'sha1 mismatch: %s has sha1 %s expected %s referenced by %s' %
1281
(record.key, sha1, item_data[1], item_data[2]))
1154
1284
def create(a_bzrdir):
1155
1285
"""Construct the current default format repository in a_bzrdir."""
1424
1576
def suspend_write_group(self):
1425
1577
raise errors.UnsuspendableWriteGroup(self)
1579
def get_missing_parent_inventories(self, check_for_missing_texts=True):
1580
"""Return the keys of missing inventory parents for revisions added in
1583
A revision is not complete if the inventory delta for that revision
1584
cannot be calculated. Therefore if the parent inventories of a
1585
revision are not present, the revision is incomplete, and e.g. cannot
1586
be streamed by a smart server. This method finds missing inventory
1587
parents for revisions added in this write group.
1589
if not self._format.supports_external_lookups:
1590
# This is only an issue for stacked repositories
1592
if not self.is_in_write_group():
1593
raise AssertionError('not in a write group')
1595
# XXX: We assume that every added revision already has its
1596
# corresponding inventory, so we only check for parent inventories that
1597
# might be missing, rather than all inventories.
1598
parents = set(self.revisions._index.get_missing_parents())
1599
parents.discard(_mod_revision.NULL_REVISION)
1600
unstacked_inventories = self.inventories._index
1601
present_inventories = unstacked_inventories.get_parent_map(
1602
key[-1:] for key in parents)
1603
parents.difference_update(present_inventories)
1604
if len(parents) == 0:
1605
# No missing parent inventories.
1607
if not check_for_missing_texts:
1608
return set(('inventories', rev_id) for (rev_id,) in parents)
1609
# Ok, now we have a list of missing inventories. But these only matter
1610
# if the inventories that reference them are missing some texts they
1611
# appear to introduce.
1612
# XXX: Texts referenced by all added inventories need to be present,
1613
# but at the moment we're only checking for texts referenced by
1614
# inventories at the graph's edge.
1615
key_deps = self.revisions._index._key_dependencies
1616
key_deps.satisfy_refs_for_keys(present_inventories)
1617
referrers = frozenset(r[0] for r in key_deps.get_referrers())
1618
file_ids = self.fileids_altered_by_revision_ids(referrers)
1619
missing_texts = set()
1620
for file_id, version_ids in file_ids.iteritems():
1621
missing_texts.update(
1622
(file_id, version_id) for version_id in version_ids)
1623
present_texts = self.texts.get_parent_map(missing_texts)
1624
missing_texts.difference_update(present_texts)
1625
if not missing_texts:
1626
# No texts are missing, so all revisions and their deltas are
1629
# Alternatively the text versions could be returned as the missing
1630
# keys, but this is likely to be less data.
1631
missing_keys = set(('inventories', rev_id) for (rev_id,) in parents)
1427
1634
def refresh_data(self):
1428
1635
"""Re-read any data needed to to synchronise with disk.
2114
2376
:param revision_ids: The expected revision ids of the inventories.
2377
:param ordering: optional ordering, e.g. 'topological'. If not
2378
specified, the order of revision_ids will be preserved (by
2379
buffering if necessary).
2115
2380
:return: An iterator of inventories.
2117
2382
if ((None in revision_ids)
2118
2383
or (_mod_revision.NULL_REVISION in revision_ids)):
2119
2384
raise ValueError('cannot get null revision inventory')
2120
return self._iter_inventories(revision_ids)
2385
return self._iter_inventories(revision_ids, ordering)
2122
def _iter_inventories(self, revision_ids):
2387
def _iter_inventories(self, revision_ids, ordering):
2123
2388
"""single-document based inventory iteration."""
2124
for text, revision_id in self._iter_inventory_xmls(revision_ids):
2389
inv_xmls = self._iter_inventory_xmls(revision_ids, ordering)
2390
for text, revision_id in inv_xmls:
2125
2391
yield self.deserialise_inventory(revision_id, text)
2127
def _iter_inventory_xmls(self, revision_ids):
2393
def _iter_inventory_xmls(self, revision_ids, ordering):
2394
if ordering is None:
2395
order_as_requested = True
2396
ordering = 'unordered'
2398
order_as_requested = False
2128
2399
keys = [(revision_id,) for revision_id in revision_ids]
2129
stream = self.inventories.get_record_stream(keys, 'unordered', True)
2402
if order_as_requested:
2403
key_iter = iter(keys)
2404
next_key = key_iter.next()
2405
stream = self.inventories.get_record_stream(keys, ordering, True)
2130
2406
text_chunks = {}
2131
2407
for record in stream:
2132
2408
if record.storage_kind != 'absent':
2133
text_chunks[record.key] = record.get_bytes_as('chunked')
2409
chunks = record.get_bytes_as('chunked')
2410
if order_as_requested:
2411
text_chunks[record.key] = chunks
2413
yield ''.join(chunks), record.key[-1]
2135
2415
raise errors.NoSuchRevision(self, record.key)
2137
chunks = text_chunks.pop(key)
2138
yield ''.join(chunks), key[-1]
2416
if order_as_requested:
2417
# Yield as many results as we can while preserving order.
2418
while next_key in text_chunks:
2419
chunks = text_chunks.pop(next_key)
2420
yield ''.join(chunks), next_key[-1]
2422
next_key = key_iter.next()
2423
except StopIteration:
2424
# We still want to fully consume the get_record_stream,
2425
# just in case it is not actually finished at this point
2140
2429
def deserialise_inventory(self, revision_id, xml):
2141
2430
"""Transform the xml into an inventory object.
3356
3724
return self.source.revision_ids_to_search_result(result_set)
3359
class InterPackRepo(InterSameDataRepository):
3360
"""Optimised code paths between Pack based repositories."""
3363
def _get_repo_format_to_test(self):
3364
from bzrlib.repofmt import pack_repo
3365
return pack_repo.RepositoryFormatKnitPack6RichRoot()
3368
def is_compatible(source, target):
3369
"""Be compatible with known Pack formats.
3371
We don't test for the stores being of specific types because that
3372
could lead to confusing results, and there is no need to be
3375
InterPackRepo does not support CHK based repositories.
3377
from bzrlib.repofmt.pack_repo import RepositoryFormatPack
3378
from bzrlib.repofmt.groupcompress_repo import RepositoryFormatCHK1
3380
are_packs = (isinstance(source._format, RepositoryFormatPack) and
3381
isinstance(target._format, RepositoryFormatPack))
3382
not_packs = (isinstance(source._format, RepositoryFormatCHK1) or
3383
isinstance(target._format, RepositoryFormatCHK1))
3384
except AttributeError:
3386
if not_packs or not are_packs:
3388
return InterRepository._same_model(source, target)
3391
def fetch(self, revision_id=None, pb=None, find_ghosts=False,
3393
"""See InterRepository.fetch()."""
3394
if (len(self.source._fallback_repositories) > 0 or
3395
len(self.target._fallback_repositories) > 0):
3396
# The pack layer is not aware of fallback repositories, so when
3397
# fetching from a stacked repository or into a stacked repository
3398
# we use the generic fetch logic which uses the VersionedFiles
3399
# attributes on repository.
3400
from bzrlib.fetch import RepoFetcher
3401
fetcher = RepoFetcher(self.target, self.source, revision_id,
3402
pb, find_ghosts, fetch_spec=fetch_spec)
3403
if fetch_spec is not None:
3404
if len(list(fetch_spec.heads)) != 1:
3405
raise AssertionError(
3406
"InterPackRepo.fetch doesn't support "
3407
"fetching multiple heads yet.")
3408
revision_id = list(fetch_spec.heads)[0]
3410
if revision_id is None:
3412
# everything to do - use pack logic
3413
# to fetch from all packs to one without
3414
# inventory parsing etc, IFF nothing to be copied is in the target.
3416
source_revision_ids = frozenset(self.source.all_revision_ids())
3417
revision_ids = source_revision_ids - \
3418
frozenset(self.target.get_parent_map(source_revision_ids))
3419
revision_keys = [(revid,) for revid in revision_ids]
3420
index = self.target._pack_collection.revision_index.combined_index
3421
present_revision_ids = set(item[1][0] for item in
3422
index.iter_entries(revision_keys))
3423
revision_ids = set(revision_ids) - present_revision_ids
3424
# implementing the TODO will involve:
3425
# - detecting when all of a pack is selected
3426
# - avoiding as much as possible pre-selection, so the
3427
# more-core routines such as create_pack_from_packs can filter in
3428
# a just-in-time fashion. (though having a HEADS list on a
3429
# repository might make this a lot easier, because we could
3430
# sensibly detect 'new revisions' without doing a full index scan.
3431
elif _mod_revision.is_null(revision_id):
3436
revision_ids = self.search_missing_revision_ids(revision_id,
3437
find_ghosts=find_ghosts).get_keys()
3438
except errors.NoSuchRevision:
3439
raise errors.InstallFailed([revision_id])
3440
if len(revision_ids) == 0:
3442
return self._pack(self.source, self.target, revision_ids)
3444
def _pack(self, source, target, revision_ids):
3445
from bzrlib.repofmt.pack_repo import Packer
3446
packs = source._pack_collection.all_packs()
3447
pack = Packer(self.target._pack_collection, packs, '.fetch',
3448
revision_ids).pack()
3449
if pack is not None:
3450
self.target._pack_collection._save_pack_names()
3451
copied_revs = pack.get_revision_count()
3452
# Trigger an autopack. This may duplicate effort as we've just done
3453
# a pack creation, but for now it is simpler to think about as
3454
# 'upload data, then repack if needed'.
3455
self.target._pack_collection.autopack()
3456
return (copied_revs, [])
3461
def search_missing_revision_ids(self, revision_id=None, find_ghosts=True):
3462
"""See InterRepository.missing_revision_ids().
3464
:param find_ghosts: Find ghosts throughout the ancestry of
3467
if not find_ghosts and revision_id is not None:
3468
return self._walk_to_common_revisions([revision_id])
3469
elif revision_id is not None:
3470
# Find ghosts: search for revisions pointing from one repository to
3471
# the other, and vice versa, anywhere in the history of revision_id.
3472
graph = self.target.get_graph(other_repository=self.source)
3473
searcher = graph._make_breadth_first_searcher([revision_id])
3477
next_revs, ghosts = searcher.next_with_ghosts()
3478
except StopIteration:
3480
if revision_id in ghosts:
3481
raise errors.NoSuchRevision(self.source, revision_id)
3482
found_ids.update(next_revs)
3483
found_ids.update(ghosts)
3484
found_ids = frozenset(found_ids)
3485
# Double query here: should be able to avoid this by changing the
3486
# graph api further.
3487
result_set = found_ids - frozenset(
3488
self.target.get_parent_map(found_ids))
3490
source_ids = self.source.all_revision_ids()
3491
# source_ids is the worst possible case we may need to pull.
3492
# now we want to filter source_ids against what we actually
3493
# have in target, but don't try to check for existence where we know
3494
# we do not have a revision as that would be pointless.
3495
target_ids = set(self.target.all_revision_ids())
3496
result_set = set(source_ids).difference(target_ids)
3497
return self.source.revision_ids_to_search_result(result_set)
3500
3727
class InterDifferingSerializer(InterRepository):
3509
3736
# This is redundant with format.check_conversion_target(), however that
3510
3737
# raises an exception, and we just want to say "False" as in we won't
3511
3738
# support converting between these formats.
3739
if 'IDS_never' in debug.debug_flags:
3512
3741
if source.supports_rich_root() and not target.supports_rich_root():
3514
3743
if (source._format.supports_tree_reference
3515
3744
and not target._format.supports_tree_reference):
3746
if target._fallback_repositories and target._format.supports_chks:
3747
# IDS doesn't know how to copy CHKs for the parent inventories it
3748
# adds to stacked repos.
3750
if 'IDS_always' in debug.debug_flags:
3752
# Only use this code path for local source and target. IDS does far
3753
# too much IO (both bandwidth and roundtrips) over a network.
3754
if not source.bzrdir.transport.base.startswith('file:///'):
3756
if not target.bzrdir.transport.base.startswith('file:///'):
3519
def _get_delta_for_revision(self, tree, parent_ids, basis_id, cache):
3760
def _get_trees(self, revision_ids, cache):
3762
for rev_id in revision_ids:
3764
possible_trees.append((rev_id, cache[rev_id]))
3766
# Not cached, but inventory might be present anyway.
3768
tree = self.source.revision_tree(rev_id)
3769
except errors.NoSuchRevision:
3770
# Nope, parent is ghost.
3773
cache[rev_id] = tree
3774
possible_trees.append((rev_id, tree))
3775
return possible_trees
3777
def _get_delta_for_revision(self, tree, parent_ids, possible_trees):
3520
3778
"""Get the best delta and base for this revision.
3522
3780
:return: (basis_id, delta)
3524
possible_trees = [(parent_id, cache[parent_id])
3525
for parent_id in parent_ids
3526
if parent_id in cache]
3527
if len(possible_trees) == 0:
3528
# There either aren't any parents, or the parents aren't in the
3529
# cache, so just use the last converted tree
3530
possible_trees.append((basis_id, cache[basis_id]))
3783
# Generate deltas against each tree, to find the shortest.
3784
texts_possibly_new_in_tree = set()
3532
3785
for basis_id, basis_tree in possible_trees:
3533
3786
delta = tree.inventory._make_delta(basis_tree.inventory)
3787
for old_path, new_path, file_id, new_entry in delta:
3788
if new_path is None:
3789
# This file_id isn't present in the new rev, so we don't
3793
# Rich roots are handled elsewhere...
3795
kind = new_entry.kind
3796
if kind != 'directory' and kind != 'file':
3797
# No text record associated with this inventory entry.
3799
# This is a directory or file that has changed somehow.
3800
texts_possibly_new_in_tree.add((file_id, new_entry.revision))
3534
3801
deltas.append((len(delta), basis_id, delta))
3536
3803
return deltas[0][1:]
3538
def _get_parent_keys(self, root_key, parent_map):
3539
"""Get the parent keys for a given root id."""
3540
root_id, rev_id = root_key
3541
# Include direct parents of the revision, but only if they used
3544
for parent_id in parent_map[rev_id]:
3545
if parent_id == _mod_revision.NULL_REVISION:
3547
if parent_id not in self._revision_id_to_root_id:
3548
# We probably didn't read this revision, go spend the
3549
# extra effort to actually check
3551
tree = self.source.revision_tree(parent_id)
3552
except errors.NoSuchRevision:
3553
# Ghost, fill out _revision_id_to_root_id in case we
3554
# encounter this again.
3555
# But set parent_root_id to None since we don't really know
3556
parent_root_id = None
3558
parent_root_id = tree.get_root_id()
3559
self._revision_id_to_root_id[parent_id] = None
3561
parent_root_id = self._revision_id_to_root_id[parent_id]
3562
if root_id == parent_root_id or parent_root_id is None:
3563
parent_keys.append((root_id, parent_id))
3564
return tuple(parent_keys)
3566
def _new_root_data_stream(self, root_keys_to_create, parent_map):
3567
for root_key in root_keys_to_create:
3568
parent_keys = self._get_parent_keys(root_key, parent_map)
3569
yield versionedfile.FulltextContentFactory(root_key,
3570
parent_keys, None, '')
3572
def _fetch_batch(self, revision_ids, basis_id, cache):
3805
def _fetch_parent_invs_for_stacking(self, parent_map, cache):
3806
"""Find all parent revisions that are absent, but for which the
3807
inventory is present, and copy those inventories.
3809
This is necessary to preserve correctness when the source is stacked
3810
without fallbacks configured. (Note that in cases like upgrade the
3811
source may be not have _fallback_repositories even though it is
3815
for parents in parent_map.values():
3816
parent_revs.update(parents)
3817
present_parents = self.source.get_parent_map(parent_revs)
3818
absent_parents = set(parent_revs).difference(present_parents)
3819
parent_invs_keys_for_stacking = self.source.inventories.get_parent_map(
3820
(rev_id,) for rev_id in absent_parents)
3821
parent_inv_ids = [key[-1] for key in parent_invs_keys_for_stacking]
3822
for parent_tree in self.source.revision_trees(parent_inv_ids):
3823
current_revision_id = parent_tree.get_revision_id()
3824
parents_parents_keys = parent_invs_keys_for_stacking[
3825
(current_revision_id,)]
3826
parents_parents = [key[-1] for key in parents_parents_keys]
3827
basis_id = _mod_revision.NULL_REVISION
3828
basis_tree = self.source.revision_tree(basis_id)
3829
delta = parent_tree.inventory._make_delta(basis_tree.inventory)
3830
self.target.add_inventory_by_delta(
3831
basis_id, delta, current_revision_id, parents_parents)
3832
cache[current_revision_id] = parent_tree
3834
def _fetch_batch(self, revision_ids, basis_id, cache, a_graph=None):
3573
3835
"""Fetch across a few revisions.
3575
3837
:param revision_ids: The revisions to copy
3576
3838
:param basis_id: The revision_id of a tree that must be in cache, used
3577
3839
as a basis for delta when no other base is available
3578
3840
:param cache: A cache of RevisionTrees that we can use.
3841
:param a_graph: A Graph object to determine the heads() of the
3842
rich-root data stream.
3579
3843
:return: The revision_id of the last converted tree. The RevisionTree
3580
3844
for it will be in cache
3587
3851
pending_deltas = []
3588
3852
pending_revisions = []
3589
3853
parent_map = self.source.get_parent_map(revision_ids)
3854
self._fetch_parent_invs_for_stacking(parent_map, cache)
3855
self.source._safe_to_return_from_cache = True
3590
3856
for tree in self.source.revision_trees(revision_ids):
3857
# Find a inventory delta for this revision.
3858
# Find text entries that need to be copied, too.
3591
3859
current_revision_id = tree.get_revision_id()
3592
3860
parent_ids = parent_map.get(current_revision_id, ())
3861
parent_trees = self._get_trees(parent_ids, cache)
3862
possible_trees = list(parent_trees)
3863
if len(possible_trees) == 0:
3864
# There either aren't any parents, or the parents are ghosts,
3865
# so just use the last converted tree.
3866
possible_trees.append((basis_id, cache[basis_id]))
3593
3867
basis_id, delta = self._get_delta_for_revision(tree, parent_ids,
3869
revision = self.source.get_revision(current_revision_id)
3870
pending_deltas.append((basis_id, delta,
3871
current_revision_id, revision.parent_ids))
3595
3872
if self._converting_to_rich_root:
3596
3873
self._revision_id_to_root_id[current_revision_id] = \
3597
3874
tree.get_root_id()
3598
# Find text entries that need to be copied
3875
# Determine which texts are in present in this revision but not in
3876
# any of the available parents.
3877
texts_possibly_new_in_tree = set()
3599
3878
for old_path, new_path, file_id, entry in delta:
3600
if new_path is not None:
3603
if not self.target.supports_rich_root():
3604
# The target doesn't support rich root, so we don't
3607
if self._converting_to_rich_root:
3608
# This can't be copied normally, we have to insert
3610
root_keys_to_create.add((file_id, entry.revision))
3612
text_keys.add((file_id, entry.revision))
3613
revision = self.source.get_revision(current_revision_id)
3614
pending_deltas.append((basis_id, delta,
3615
current_revision_id, revision.parent_ids))
3879
if new_path is None:
3880
# This file_id isn't present in the new rev
3884
if not self.target.supports_rich_root():
3885
# The target doesn't support rich root, so we don't
3888
if self._converting_to_rich_root:
3889
# This can't be copied normally, we have to insert
3891
root_keys_to_create.add((file_id, entry.revision))
3894
texts_possibly_new_in_tree.add((file_id, entry.revision))
3895
for basis_id, basis_tree in possible_trees:
3896
basis_inv = basis_tree.inventory
3897
for file_key in list(texts_possibly_new_in_tree):
3898
file_id, file_revision = file_key
3900
entry = basis_inv[file_id]
3901
except errors.NoSuchId:
3903
if entry.revision == file_revision:
3904
texts_possibly_new_in_tree.remove(file_key)
3905
text_keys.update(texts_possibly_new_in_tree)
3616
3906
pending_revisions.append(revision)
3617
3907
cache[current_revision_id] = tree
3618
3908
basis_id = current_revision_id
3909
self.source._safe_to_return_from_cache = False
3619
3910
# Copy file texts
3620
3911
from_texts = self.source.texts
3621
3912
to_texts = self.target.texts
3622
3913
if root_keys_to_create:
3623
root_stream = self._new_root_data_stream(root_keys_to_create,
3914
root_stream = _mod_fetch._new_root_data_stream(
3915
root_keys_to_create, self._revision_id_to_root_id, parent_map,
3916
self.source, graph=a_graph)
3625
3917
to_texts.insert_record_stream(root_stream)
3626
3918
to_texts.insert_record_stream(from_texts.get_record_stream(
3627
3919
text_keys, self.target._format._fetch_order,
4020
4375
# missing keys can handle suspending a write group).
4021
4376
write_group_tokens = self.target_repo.suspend_write_group()
4022
4377
return write_group_tokens, missing_keys
4023
self.target_repo.commit_write_group()
4378
hint = self.target_repo.commit_write_group()
4379
if (to_serializer != src_serializer and
4380
self.target_repo._format.pack_compresses):
4381
self.target_repo.pack(hint=hint)
4024
4382
return [], set()
4026
def _extract_and_insert_inventories(self, substream, serializer):
4384
def _extract_and_insert_inventory_deltas(self, substream, serializer):
4385
target_rich_root = self.target_repo._format.rich_root_data
4386
target_tree_refs = self.target_repo._format.supports_tree_reference
4387
for record in substream:
4388
# Insert the delta directly
4389
inventory_delta_bytes = record.get_bytes_as('fulltext')
4390
deserialiser = inventory_delta.InventoryDeltaDeserializer()
4392
parse_result = deserialiser.parse_text_bytes(
4393
inventory_delta_bytes)
4394
except inventory_delta.IncompatibleInventoryDelta, err:
4395
trace.mutter("Incompatible delta: %s", err.msg)
4396
raise errors.IncompatibleRevision(self.target_repo._format)
4397
basis_id, new_id, rich_root, tree_refs, inv_delta = parse_result
4398
revision_id = new_id
4399
parents = [key[0] for key in record.parents]
4400
self.target_repo.add_inventory_by_delta(
4401
basis_id, inv_delta, revision_id, parents)
4403
def _extract_and_insert_inventories(self, substream, serializer,
4027
4405
"""Generate a new inventory versionedfile in target, converting data.
4029
4407
The inventory is retrieved from the source, (deserializing it), and
4030
4408
stored in the target (reserializing it in a different format).
4410
target_rich_root = self.target_repo._format.rich_root_data
4411
target_tree_refs = self.target_repo._format.supports_tree_reference
4032
4412
for record in substream:
4413
# It's not a delta, so it must be a fulltext in the source
4414
# serializer's format.
4033
4415
bytes = record.get_bytes_as('fulltext')
4034
4416
revision_id = record.key[0]
4035
4417
inv = serializer.read_inventory_from_string(bytes, revision_id)
4036
4418
parents = [key[0] for key in record.parents]
4037
4419
self.target_repo.add_inventory(revision_id, inv, parents)
4420
# No need to keep holding this full inv in memory when the rest of
4421
# the substream is likely to be all deltas.
4039
4424
def _extract_and_insert_revisions(self, substream, serializer):
4040
4425
for record in substream:
4186
4588
return (not self.from_repository._format.rich_root_data and
4187
4589
self.to_format.rich_root_data)
4189
def _get_inventory_stream(self, revision_ids):
4591
def _get_inventory_stream(self, revision_ids, missing=False):
4190
4592
from_format = self.from_repository._format
4191
if (from_format.supports_chks and self.to_format.supports_chks
4192
and (from_format._serializer == self.to_format._serializer)):
4193
# Both sides support chks, and they use the same serializer, so it
4194
# is safe to transmit the chk pages and inventory pages across
4196
return self._get_chk_inventory_stream(revision_ids)
4197
elif (not from_format.supports_chks):
4198
# Source repository doesn't support chks. So we can transmit the
4199
# inventories 'as-is' and either they are just accepted on the
4200
# target, or the Sink will properly convert it.
4201
return self._get_simple_inventory_stream(revision_ids)
4593
if (from_format.supports_chks and self.to_format.supports_chks and
4594
from_format.network_name() == self.to_format.network_name()):
4595
raise AssertionError(
4596
"this case should be handled by GroupCHKStreamSource")
4597
elif 'forceinvdeltas' in debug.debug_flags:
4598
return self._get_convertable_inventory_stream(revision_ids,
4599
delta_versus_null=missing)
4600
elif from_format.network_name() == self.to_format.network_name():
4602
return self._get_simple_inventory_stream(revision_ids,
4604
elif (not from_format.supports_chks and not self.to_format.supports_chks
4605
and from_format._serializer == self.to_format._serializer):
4606
# Essentially the same format.
4607
return self._get_simple_inventory_stream(revision_ids,
4203
# XXX: Hack to make not-chk->chk fetch: copy the inventories as
4204
# inventories. Note that this should probably be done somehow
4205
# as part of bzrlib.repository.StreamSink. Except JAM couldn't
4206
# figure out how a non-chk repository could possibly handle
4207
# deserializing an inventory stream from a chk repo, as it
4208
# doesn't have a way to understand individual pages.
4209
return self._get_convertable_inventory_stream(revision_ids)
4610
# Any time we switch serializations, we want to use an
4611
# inventory-delta based approach.
4612
return self._get_convertable_inventory_stream(revision_ids,
4613
delta_versus_null=missing)
4211
def _get_simple_inventory_stream(self, revision_ids):
4615
def _get_simple_inventory_stream(self, revision_ids, missing=False):
4616
# NB: This currently reopens the inventory weave in source;
4617
# using a single stream interface instead would avoid this.
4212
4618
from_weave = self.from_repository.inventories
4620
delta_closure = True
4622
delta_closure = not self.delta_on_metadata()
4213
4623
yield ('inventories', from_weave.get_record_stream(
4214
4624
[(rev_id,) for rev_id in revision_ids],
4215
self.inventory_fetch_order(),
4216
not self.delta_on_metadata()))
4218
def _get_chk_inventory_stream(self, revision_ids):
4219
"""Fetch the inventory texts, along with the associated chk maps."""
4220
# We want an inventory outside of the search set, so that we can filter
4221
# out uninteresting chk pages. For now we use
4222
# _find_revision_outside_set, but if we had a Search with cut_revs, we
4223
# could use that instead.
4224
start_rev_id = self.from_repository._find_revision_outside_set(
4226
start_rev_key = (start_rev_id,)
4227
inv_keys_to_fetch = [(rev_id,) for rev_id in revision_ids]
4228
if start_rev_id != _mod_revision.NULL_REVISION:
4229
inv_keys_to_fetch.append((start_rev_id,))
4230
# Any repo that supports chk_bytes must also support out-of-order
4231
# insertion. At least, that is how we expect it to work
4232
# We use get_record_stream instead of iter_inventories because we want
4233
# to be able to insert the stream as well. We could instead fetch
4234
# allowing deltas, and then iter_inventories, but we don't know whether
4235
# source or target is more 'local' anway.
4236
inv_stream = self.from_repository.inventories.get_record_stream(
4237
inv_keys_to_fetch, 'unordered',
4238
True) # We need them as full-texts so we can find their references
4239
uninteresting_chk_roots = set()
4240
interesting_chk_roots = set()
4241
def filter_inv_stream(inv_stream):
4242
for idx, record in enumerate(inv_stream):
4243
### child_pb.update('fetch inv', idx, len(inv_keys_to_fetch))
4244
bytes = record.get_bytes_as('fulltext')
4245
chk_inv = inventory.CHKInventory.deserialise(
4246
self.from_repository.chk_bytes, bytes, record.key)
4247
if record.key == start_rev_key:
4248
uninteresting_chk_roots.add(chk_inv.id_to_entry.key())
4249
p_id_map = chk_inv.parent_id_basename_to_file_id
4250
if p_id_map is not None:
4251
uninteresting_chk_roots.add(p_id_map.key())
4254
interesting_chk_roots.add(chk_inv.id_to_entry.key())
4255
p_id_map = chk_inv.parent_id_basename_to_file_id
4256
if p_id_map is not None:
4257
interesting_chk_roots.add(p_id_map.key())
4258
### pb.update('fetch inventory', 0, 2)
4259
yield ('inventories', filter_inv_stream(inv_stream))
4260
# Now that we have worked out all of the interesting root nodes, grab
4261
# all of the interesting pages and insert them
4262
### pb.update('fetch inventory', 1, 2)
4263
interesting = chk_map.iter_interesting_nodes(
4264
self.from_repository.chk_bytes, interesting_chk_roots,
4265
uninteresting_chk_roots)
4266
def to_stream_adapter():
4267
"""Adapt the iter_interesting_nodes result to a single stream.
4269
iter_interesting_nodes returns records as it processes them, along
4270
with keys. However, we only want to return the records themselves.
4272
for record, items in interesting:
4273
if record is not None:
4275
# XXX: We could instead call get_record_stream(records.keys())
4276
# ATM, this will always insert the records as fulltexts, and
4277
# requires that you can hang on to records once you have gone
4278
# on to the next one. Further, it causes the target to
4279
# recompress the data. Testing shows it to be faster than
4280
# requesting the records again, though.
4281
yield ('chk_bytes', to_stream_adapter())
4282
### pb.update('fetch inventory', 2, 2)
4284
def _get_convertable_inventory_stream(self, revision_ids):
4285
# XXX: One of source or target is using chks, and they don't have
4286
# compatible serializations. The StreamSink code expects to be
4287
# able to convert on the target, so we need to put
4288
# bytes-on-the-wire that can be converted
4289
yield ('inventories', self._stream_invs_as_fulltexts(revision_ids))
4291
def _stream_invs_as_fulltexts(self, revision_ids):
4625
self.inventory_fetch_order(), delta_closure))
4627
def _get_convertable_inventory_stream(self, revision_ids,
4628
delta_versus_null=False):
4629
# The source is using CHKs, but the target either doesn't or it has a
4630
# different serializer. The StreamSink code expects to be able to
4631
# convert on the target, so we need to put bytes-on-the-wire that can
4632
# be converted. That means inventory deltas (if the remote is <1.19,
4633
# RemoteStreamSink will fallback to VFS to insert the deltas).
4634
yield ('inventory-deltas',
4635
self._stream_invs_as_deltas(revision_ids,
4636
delta_versus_null=delta_versus_null))
4638
def _stream_invs_as_deltas(self, revision_ids, delta_versus_null=False):
4639
"""Return a stream of inventory-deltas for the given rev ids.
4641
:param revision_ids: The list of inventories to transmit
4642
:param delta_versus_null: Don't try to find a minimal delta for this
4643
entry, instead compute the delta versus the NULL_REVISION. This
4644
effectively streams a complete inventory. Used for stuff like
4645
filling in missing parents, etc.
4292
4647
from_repo = self.from_repository
4293
from_serializer = from_repo._format._serializer
4294
4648
revision_keys = [(rev_id,) for rev_id in revision_ids]
4295
4649
parent_map = from_repo.inventories.get_parent_map(revision_keys)
4296
for inv in self.from_repository.iter_inventories(revision_ids):
4297
# XXX: This is a bit hackish, but it works. Basically,
4298
# CHKSerializer 'accidentally' supports
4299
# read/write_inventory_to_string, even though that is never
4300
# the format that is stored on disk. It *does* give us a
4301
# single string representation for an inventory, so live with
4303
# This would be far better if we had a 'serialized inventory
4304
# delta' form. Then we could use 'inventory._make_delta', and
4305
# transmit that. This would both be faster to generate, and
4306
# result in fewer bytes-on-the-wire.
4307
as_bytes = from_serializer.write_inventory_to_string(inv)
4650
# XXX: possibly repos could implement a more efficient iter_inv_deltas
4652
inventories = self.from_repository.iter_inventories(
4653
revision_ids, 'topological')
4654
format = from_repo._format
4655
invs_sent_so_far = set([_mod_revision.NULL_REVISION])
4656
inventory_cache = lru_cache.LRUCache(50)
4657
null_inventory = from_repo.revision_tree(
4658
_mod_revision.NULL_REVISION).inventory
4659
# XXX: ideally the rich-root/tree-refs flags would be per-revision, not
4660
# per-repo (e.g. streaming a non-rich-root revision out of a rich-root
4661
# repo back into a non-rich-root repo ought to be allowed)
4662
serializer = inventory_delta.InventoryDeltaSerializer(
4663
versioned_root=format.rich_root_data,
4664
tree_references=format.supports_tree_reference)
4665
for inv in inventories:
4308
4666
key = (inv.revision_id,)
4309
4667
parent_keys = parent_map.get(key, ())
4669
if not delta_versus_null and parent_keys:
4670
# The caller did not ask for complete inventories and we have
4671
# some parents that we can delta against. Make a delta against
4672
# each parent so that we can find the smallest.
4673
parent_ids = [parent_key[0] for parent_key in parent_keys]
4674
for parent_id in parent_ids:
4675
if parent_id not in invs_sent_so_far:
4676
# We don't know that the remote side has this basis, so
4679
if parent_id == _mod_revision.NULL_REVISION:
4680
parent_inv = null_inventory
4682
parent_inv = inventory_cache.get(parent_id, None)
4683
if parent_inv is None:
4684
parent_inv = from_repo.get_inventory(parent_id)
4685
candidate_delta = inv._make_delta(parent_inv)
4686
if (delta is None or
4687
len(delta) > len(candidate_delta)):
4688
delta = candidate_delta
4689
basis_id = parent_id
4691
# Either none of the parents ended up being suitable, or we
4692
# were asked to delta against NULL
4693
basis_id = _mod_revision.NULL_REVISION
4694
delta = inv._make_delta(null_inventory)
4695
invs_sent_so_far.add(inv.revision_id)
4696
inventory_cache[inv.revision_id] = inv
4697
delta_serialized = ''.join(
4698
serializer.delta_to_lines(basis_id, key[-1], delta))
4310
4699
yield versionedfile.FulltextContentFactory(
4311
key, parent_keys, None, as_bytes)
4700
key, parent_keys, None, delta_serialized)
4703
def _iter_for_revno(repo, partial_history_cache, stop_index=None,
4704
stop_revision=None):
4705
"""Extend the partial history to include a given index
4707
If a stop_index is supplied, stop when that index has been reached.
4708
If a stop_revision is supplied, stop when that revision is
4709
encountered. Otherwise, stop when the beginning of history is
4712
:param stop_index: The index which should be present. When it is
4713
present, history extension will stop.
4714
:param stop_revision: The revision id which should be present. When
4715
it is encountered, history extension will stop.
4717
start_revision = partial_history_cache[-1]
4718
iterator = repo.iter_reverse_revision_history(start_revision)
4720
#skip the last revision in the list
4723
if (stop_index is not None and
4724
len(partial_history_cache) > stop_index):
4726
if partial_history_cache[-1] == stop_revision:
4728
revision_id = iterator.next()
4729
partial_history_cache.append(revision_id)
4730
except StopIteration:
added
removed
bzrlib/repository.py
